xenonshare/NanoMend-1.5-Ultimate

🛠️ NanoMend 1.5 Ultimate (GGUF)

NanoMend-1.5-Ultimate is a fine-tuned, CPU-optimized 1.5B parameter language model specialized in Python AST-aware logic trap detection, code repair, and performance optimization.

Engineered specifically for local, privacy-first developer tooling (such as VS Code extensions), it pairs deterministically with AST parsers to diagnose infinite loops, resource leaks, semantic intent errors, O(N²) performance issues, and static index errors without sending code to cloud APIs.

---

🌟 Key Features

• Quantization: CPU-friendly GGUF quantization format (llama.cpp compatible).
• Prompt Format: ChatML (<|im_start|>system...)
• Context Length: 2048 tokens
• Specialization: Python Code Repair, AST Trap Resolution, Performance Refactoring.
• Privacy: 100% offline, zero external telemetry.

---

⚡ Quickstart Usage (Python / llama-cpp-python)

You can run NanoMend-1.5-Ultimate locally using Python in just a few lines:

pip install llama-cpp-python huggingface_hub

from huggingface_hub import hf_hub_download
from llama_cpp import Llama

# 1. Download model from Hugging Face Hub
model_path = hf_hub_download(
    repo_id="xenonshare/NanoMend-1.5-Ultimate",
    filename="nanomend-ultimate-1.5b.gguf"
)

# 2. Load model into CPU memory
llm = Llama(
    model_path=model_path,
    n_ctx=2048,
    n_threads=8,
    verbose=False
)

# 3. Format Prompt (ChatML Format)
system_prompt = "You are a Senior Python Developer. The user's code has a CRITICAL error: INTENT ERROR. Function 'add' implies addition but uses multiplication (*). Rewrite the code to fix this. Output ONLY the fixed python code."
code_snippet = """def add(a, b):
    return a * b
"""

prompt = f"<|im_start|>system\n{system_prompt}<|im_end|>\n<|im_start|>user\n{code_snippet}<|im_end|>\n<|im_start|>assistant\n"

# 4. Generate Repair
output = llm(
    prompt,
    max_tokens=256,
    stop=["<|im_end|>"],
    temperature=0.1
)

print(output['choices'][0]['text'])

---

💬 Prompt Template

NanoMend uses the standard ChatML format:

<|im_start|>system
You are a Senior Python Developer. The user's code has a CRITICAL error: {AST_WARNING}. Rewrite the code to fix this. Output ONLY the fixed python code.<|im_end|>
<|im_start|>user
{PYTHON_CODE}<|im_end|>
<|im_start|>assistant

---

🦙 Running with Ollama

You can import this model into Ollama by creating a Modelfile:

FROM ./nanomend-ultimate-1.5b.gguf

TEMPLATE """<|im_start|>system
{{ .System }}<|im_end|>
<|im_start|>user
{{ .Prompt }}<|im_end|>
<|im_start|>assistant
"""

PARAMETER stop "<|im_end|>"
PARAMETER temperature 0.1

Run in terminal:

ollama create nanomend -f Modelfile
ollama run nanomend "Fix this loop: while x > 0: print(x)"

---

📋 Evaluation & Scope

NanoMend-1.5-Ultimate is fine-tuned to resolve the 12 primary Python logic traps:

1. Infinite Loops (Unmutated control variables)
2. Type Safety Mismatches
3. Resource Leaks (open() without with or .close())
4. Security Risks (Hardcoded credentials)
5. $O(N^2)$ Nested Loops to $O(N)$ Hash Lookups
6. Semantic Intent (e.g. multiply using +, is_even returning odd)
7. Void Function Assignment (return None)
8. Index Out of Bounds
9. Off-by-one average calculations

---

📄 License

MIT License. Free for commercial and open-source use.